Hi Lukas,

TF binding sites (Cis -regulatory elements) are spread all over the upstream and even in downstream of genes. The upstream region specifically considered as "Promoter region". But there is no such specific guide-rule for considering the length of a DNA as promoter. Determining PROMOTER itself is a research challenge of its own. What researchers do is to consider an average length of the upstream as general promoter of genes. Now, the general length also differs species to species as well as depending on the criteria of findings. Like, in our case, we considered 200 bp upstream of TSS as proximal promoter (in Plants like Rice and Arabidopsis) to find basal TF binding sites; where as, 1000 bp upstream as general promoter regions. Al though in databases it was mentioned as 3 kb upstream as promoter (rice). In some cases I found 1 kb upstream + 200 bp downstream has been considered as promoter. In human promoter regions are considered as much as 10 kb upstream of TSS also. One thing we need to keep in our mind that the promoter region (considered by us) of one gene should not overlap the coding region of the adjacent/another genes.

I can try to provide more precise answer if you share the details of the genes and purpose of findings.

Best regards

Arindam.

Hi Arindam,

Thank you for your answer! This is already helpful and goes along with my intuition that there is no "one-fits-all" solution. To give you a better understanding of my dataset: We raised zebrafish in different conditions (basically enrichment settings vs adverse conditions) and are interested in the effect on brain development. The search for common motifs is one part of the analysis where we are simply interested in potential common regulatory modules. If one or several motifs correlate strongly with differential expression in several genes in our set, this might help us to differentiate between true biological insight and random noise which is rather common in huge datasets as obtained in NGS.

So would it be a sensible approach to probe different lengths of upstream sequences? Say, 200 bp as proximal promoter, 1 kb as general region, 3 kb as extended region of interest? As you state, distance from the TSS is not necessarily an indicator of importance, motifs can be rather remote and still matter a lot. Or should I just figure out the mean length yielding positive results in promoter cloning approaches in zebrafish and go with that?

Another thing you mention is that promoter regions and adjacent genes should not overlap. Why is that the case? Sure, it would vastly complicate the genetic architecture and our analysis but are we certain that there are no instances in which genetic motifs within a gene A can influence the expression of another gene B?

All the best,

Lukas

Hi Lukas,

Yes, these decisions are often fairly arbitrary, and there will be no "one size fits all".

For me, the best way to do this analysis would be to combine it with a functional measurement of promoter architecture such as chromatin accessibility (ATAC-seq, DNase-seq) or ChIP-seq (H3K4me3, H3K27ac). For example, I would identify peaks in the chromatin-based dataset to provide a set of intervals that represent the functionally open/active regulatory elements throughout the genome. Then intersect these with your promoter/TSS set, linking the interval coordinates to the gene ID. Then you just take your "functional" intervals of your differentially expressed genes and unchanged genes and use them with your desired package for TF motif analysis. Obviously this depends on the chromatin-based datasets available to you, but I think this gets around a lot of the arbitrary cutoffs and distances you seem to be struggling with.

Cheers,

Hamish

Hi Lukas,

Your project is quite interesting. I would like to start with your statement: " If one or several motifs correlate strongly with differential expression in several genes in our set, this might help us to differentiate between true biological insight and random noise which is rather common in huge datasets as obtained in NGS." As you are not finding any specific motif (or basic TF binding sites), you can take a general length (I don't know the length for zebra fish) for motif scan. More you can do is to split the promoter in different windows (say 100 bp each; so 10 windows for 1 kb upstream; Please find our PLoS ONE 10(9):e0137295 · September 2015 paper for methodology). This will give you a precise view about the position wise occurrence frequency of different motifs.

One important thing is - only the presence of any motif does not prove the functionality of the respective TFs. Motifs are the blue-print of TF binding and not the actual binding. In computational study we generally need to find some relation/co-relation in between the occurrence of motifs with some other factors. For example, in our case we have shown the statistical enrichment of some specific motifs in some tissue-specif as well as condition-specific diff up-regulated gene sets. And further we showed that the over-represented motifs are not by chance occurrence (by statistical score on negative control of background data).

Next, "Another thing you mention is that promoter regions and adjacent genes should not overlap."

We know that the inter-genic regions act as promoter due to some biological reasons. So, avoiding adjacent gene (coding) regions might help us to reduce false positives (findings motif occurrences in the coding regions which actually are not TF binding sites but only similar sequences.)

Hope this will help.

Best regards,

Arindam.

In addition you may want to mask repeats - could lead to a loss of real TF binding sites but may eliminate false positives.

Thanks to Hamish, Arindam, and Ross for your answers! I will definitely try to dissect the upstream regions in smaller windows as suggested by Arindam and see what I find there (the paper you pointed me at is very interesting!). Masking and complementing with ChIP-Seq data seems also promising and I will try and implement it into my pipeline. Thanks a lot!